diff --git a/setup.py b/setup.py
index 70d46cc..5f26f88 100644
--- a/setup.py
+++ b/setup.py
@@ -3,7 +3,7 @@
 setup(
   name = 'x-clip',
   packages = find_packages(exclude=[]),
-  version = '0.0.3',
+  version = '0.0.4',
   license='MIT',
   description = 'X-CLIP',
   author = 'Phil Wang',
diff --git a/x_clip/x_clip.py b/x_clip/x_clip.py
index 7cc56bb..9cd3be0 100644
--- a/x_clip/x_clip.py
+++ b/x_clip/x_clip.py
@@ -324,7 +324,7 @@ def forward(
             image_to_text = text_to_image.t()
 
             if self.extra_latent_projection:
-                image_to_text = einsum('t d, i d -> t i', text_latents_extra, image_latents_extra) * temp
+                image_to_text = einsum('t d, i d -> i t', text_latents_extra, image_latents_extra) * temp
 
         # calculate loss